HR: 17:00h
AN: B24B-05 [Abstracts]
TI: Global sensitivity analysis of Leaf-Canopy radiative transfer Model for analysis and quantification of uncertainties in remote sensed data product generation
AU: * Furfaro, R
EM: robertof@email.arizona.edu
AF: Aerospace and Mechanical Engineering Department, Univerisity of Arizona, 1130 N.
mountain, tucson, AZ 85721, United States
AU: Morris, R D
EM: robin.morris@gmail.com
AF: USRA-RIACS, 444 Castro St, Suite 320, Mountain View, CA 94041, United States
AU: Kottas, A
EM: thanos@soe,ucsc.edu
AF: Department of Applied Mathematics and Statistics, University of California, 1156 High
Street, Santa Cruz, CA 95064, United States
AU: Taddy, M
EM: taddy@soe.ecsc.edu
AF: Department of Applied Mathematics and Statistics, University of California, 1156 High
Street, Santa Cruz, CA 95064, United States
AU: Ganapol, B D
EM: ganapol@cowboy.ame.arizona.edu
AF: Aerospace and Mechanical Engineering Department, Univerisity of Arizona, 1130 N.
mountain, tucson, AZ 85721, United States
AB:
Analyzing, quantifying and reporting the uncertainty in remote sensed data
products is critical for our understanding of Earth's coupled system. It is
the only way in which the uncertainty of further analyses using these data
products as inputs can be quantified. Analyzing the source of the data product
uncertainties can identify where the models must be improved, or where better
input information must be obtained. Here we focus on developing a probabilistic
framework for analysis of uncertainties occurring when satellite data (e.g.,
MODIS) are employed to retrieve biophysical properties of vegetation. Indeed,
the process of remotely estimating vegetation properties involves inverting a
Radiative Transfer Model (RTM), as in the case of the MOD15 algorithm where
seven atmospherically corrected reflectance factors are ingested and compared
to a set of computed, RTM-based, reflectances (look-up table) to infer the Leaf
Area Index (LAI). Since inversion is generally ill-conditioned, and since a-priori
information is important in constraining the inverse model, sensitivity analysis
plays a key role in defining which parameters have the greatest impact to the
computed observation. We develop a framework to perform global sensitivity
analysis, i.e., to determine how the output changes as all inputs vary continuously.
We used a coupled Leaf-Canopy radiative transfer Model (LCM) to approximate the
functional relationship between the observed reflectance and vegetation biophysical
parameters. LCM was designed to study the feasibility of detecting leaf/canopy
biochemistry using remote sensed observations and has the unique capability to
include leaf biochemistry (e.g., chlorophyll, water, lignin, protein) as input
parameters. The influence of LCM input parameters (including canopy morphological
and biochemical parameters) on the hemispherical reflectance is captured by
computing the "main effects", which give information about the influence of each
input, and the "sensitivity indices", i.e., the expected amount by which the
uncertainty in the output is reduced if the true value of a specific input was
known. Since RTMs are generally computationally expensive, we develop a Gaussian
Process (GP) statistical model to approximate the LCM output surface. Once the GP
parameters are estimated (using simulated data from the LCM model), the computation
of main effects and sensitivity indices is straightforward. Using this approach,
we were able to quantify the importance of LCM input parameters for each of the
eight wavelengths centered on the MODIS bands commonly used to observe vegetation
(visible and Near-Infrared or NIR). We found that chlorophyll dominates the visible
region while LAI has the largest effect on the NIR. Surprisingly, we also found that
lignin is sensitive in short-wave infrared (1640nm and 2310nm) where it is the main
contributor to the overall reflectance. These results indicate the feasibility and
utility of probabilistic uncertainty and sensitivity analysis for RTMs. The
developed methodology can be used both to improve RTMs for better measurement
predictions, and to guide the data collection or land cover to reduce the level of
uncertainty.
DE: 0430 Computational methods and data processing
DE: 0480 Remote sensing
SC: Biogeosciences [B]
MN: 2007 Fall Meeting